Magnetic-Field-Based Position Sensing Using Machine Learning

Abstract

Magnetic fields are widely used in short-range wireless applications such as sensor systems and communication systems. To further exploit the potential of such systems that use magnetic fields, we investigated their applicability to position sensing of a mobile device that generates these fields. The principle involves estimating the position of the device via an analysis of the data detected by multiple magnetic-field sensors located around the target space. In this study, we used machine learning to analyze the sensor data, which were obtained by numerical calculations. The results indicated that machine learning effectively estimated the position of the mobile devices. Based on our simulations, the error of the position estimated with the machine-learning approach was within 10 cm in a 2×2×2-m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> cubic space for 73% of all the cases of mobile-device states. The estimation accuracy exceeded that obtained with a conventional optimizing approach. Furthermore, the estimation accuracy obtained with the machine learning approach was maintained for the signal-to-noise-ratio higher than 30 dB. It was also shown that the degradation of the estimation accuracy caused by a sensor-location shift can be restored by learning with training data for the shifted sensor location. The computational speed of the machine learning approach is 30 times faster than that of the conventional one. The results significantly support the applicability of magnetic-field-based systems for real-time tracking of moving persons and objects.